base_model:

• JonusNattapong/LFM3-Omni-2B-abliterated
  pipeline_tag: text-generation
  license: other
  license_name: lfm1.0
  license_link: LICENSE
  tags:
• gguf
• llama.cpp
• lfm3
• lfm3-omni
• abliterated
• tool-calling
• custom_code
  language:
• en
• th

LFM3-Omni-2B-abliterated — GGUF (v4.2)

GGUF build of
JonusNattapong/LFM3-Omni-2B-abliterated
with the v4.2 coding/tool-use LoRA merged into the BF16 weights before
conversion.

Files

No Q4 quant of the v4.2 merge is published. The official
LFM3-Omni-2B-Q4_K_M.gguf is the non-abliterated, non-v4.2 base model and
is deliberately not redistributed here.

Read this before using it

• The Q8_0 file is the text-backbone conversion of the merged model. The
  converter exports the language model; it does not contain every BF16 Omni
  tensor. The complete merged checkpoint is in the
  -BF16
  repository.
• The projector is the official LiquidAI file. It loads alongside the Q8_0
  text model, but image/audio/video inference with the v4.2 text weights was
  not verified.
• Verified: the Q8_0 file loads and produces text with llama-server
  (build 11953, commit a7af149a2) on CPU. No GPU offload benchmark exists.
• Merged weights are not numerically validated against the adapter applied
  at runtime; only successful loading and text generation were checked.
• No benchmark result is published for these weights (the pilot evaluation was
  run on a different adapter revision).
• Tool calls come out as Liquid XML
  (<tool_call><function=NAME><parameter=K>V</parameter></function></tool_call>);
  a harness must execute them.

Example

llama-server \
  --model LFM3-Omni-2B-abliterated-v4.2-Q8_0.gguf \
  --mmproj mmproj-LFM3-Omni-2B-Q4_K_M.gguf \
  --ctx-size 8192 --host 127.0.0.1 --port 8080

License

LFM Open License v1.0 from the base model; a copy is included as
LICENSE. Modified files must be marked as changed, and the
license and attribution notices must be redistributed with any copy.
The projector file is redistributed unmodified from LiquidAI.